AI agents need three types of memory: short-term for conversation history, long-term for learned preferences, and reasoning for decision traces. With only one or two memory types, key details from early exchanges are lost, leading to context rot. Agents also pay to resend tokens they no longer need, causing context waste.

Production systems demand provenance from question to answer. The open-source package, neo4j-agent-memory, is the first of its kind to deliver all three memory types your AI agents need. Backed by Neo4j’s graph database, neo4j-agent-memory enables you to build context graphs with causal chains that capture the “why” behind decisions. From multi-hop traversal across entities to cross-conversation knowledge persistence, graph-native memory achieves what vector stores alone can’t.

Join us on April 7 to learn how to get started with neo4j-agent-memory with just one pip install, and seamlessly integrate it with modern frameworks like PydanticAI, LangChain, LlamaIndex, Google ADK, Microsoft Agent Framework, and AWS Strands. 

You’ll see live demos exploring:

• A financial services context graph that uses decision and reasoning traces captured throughout the enterprise to evaluate credit decisions, identify fraud, and provide customer 360 analysis

• A context graph that includes 300+ podcast episodes and 19 agent tools, featuring conversation memory and reasoning reuse